MobileNetV3与互易点损失函数相结合的雷达波形开集识别  

Radar Waveform Open-Set Recognition Based on MobileNetV3 andReciprocal Point Loss

作　　者：刘志林 王晋东[1] 李银龙 冯蕴天 王彬 LIU Zhilin;WANG Jindong;LI Yinlong;FENG Yuntian;WANG Bin(Information Engineering University,Zhengzhou 450001,China;Unit 63893,Luoyang 471000,China)

机构地区：[1]信息工程大学,河南郑州450001 [2]63893部队,河南洛阳471000

出　　处：《信息工程大学学报》2024年第6期631-638,共8页Journal of Information Engineering University

基　　金：电子信息系统复杂电磁环境效应国家重点实验室基金(CEMEE2022Z0104A)。

摘　　要：基于深度学习的雷达波形识别方法通常假定待识别信号属于一个已知并且种类有限的集合,但在实际场景中可能存在大量未知雷达信号,导致此类闭集识别方法难以适用。针对此问题,提出一种MobileNetV3与互易点损失函数相结合的开集识别方法。利用神经网络提取信号时频图像的高维特征向量,通过特征向量和互易点的距离来衡量已知信号和未知信号之间的差异,使模型在正确识别已知信号波形的同时也能对未知信号波形进行判别,实现雷达波形的开集识别功能。实验结果表明,在6~15 dB的信噪比范围内,该方法对已知信号波形的识别准确率接近100%,对未知信号波形的判别准确率达到90%以上。In the radar waveform recognition method based on deep learning,it is usually assumes that the signals to be recognized belong to a known and limited set.However,in practical scenarios,there may be a large number of unknown radar signals during the recognition phase,making such closed-set recognition methods difficult to apply.To address this issue,a novel open-set recognition method based on Mobile-NetV3 and reciprocal point loss is proposed.A neural network is used to extract high-dimensional feature vectors of signal time-frequency maps.The distance between the feature vector and the reciprocal point is used to measure the difference between known and unknown signals,enabling the model to correctly rec-ognize known signal waveforms while also discriminating against unknown signal waveforms,thereby achieving open-set recognition of radar waveforms.Experimental results show that within a signal-to-noise ratio range of 6 to 15 dB,the recognition accuracy for known signal waveforms achieves nearly 100%,and the discrimination accuracy for unknown signal waveforms reaches over 90%by this method.

关 键 词：雷达信号 波形识别 崔-威廉斯分布 MobileNetV3网络 互易点 

分 类 号：TN957.51[电子电信—信号与信息处理]